Brent Berlin and Paul Kay. 1999 (1969). CSLI Publications: Leland Stanford University.
From the preface to the paperback edition:
"(1) there are substantive universal constraints on the shape of basic color lexicons -- systems of color naming do not vary randomly or carpiciously across langauges but are constrained to a small number of possible types; and (2) basic color lexicons change type over time by adding basic color terms in a highly constrained, though not mechanically predictable manner" [p. v] This is a substantial revision of B&K's original thesis, backing off the rather deterministic model of color-term evolution proposed in this volume.
"H.C.Conklin (1955) has shown ... that Hanuno/o 'color' words in fact encode a great deal of non-colorimetric information. The essentially methodological point made in such studies has been frequently misinterpreted by anthropologists and linguists as an argument against the existence of semantic universals." (as does Bradley). The article cited is "Hanuno/o color categories." Southwestern Journal of Anthropology 11:339-344.
This book surveys 98 different languages from different families.
Used 329 Munsell Color chips arranged on a board; asked participants to list basic colors, mark focal colors (best example of the color) and mark boundaries of the color. Boundaries were much more difficult for participants to mark, and results were not reliable. Foci were comparatively easy. [ p. 13]
"category boundaries proved to be so unreliable, even for an individual informant, that they have been accorded a relatively minor place in the analysis." 
Criteria for BCTs: [quoted from page 6]: ******
(i) It is monolexemic; that is, its meaning is not predictable from the meaning of its parts (cf. Conklin 1962)....
(ii) It signification is not included in that of any other color term....
(iii) Its application must not be restricted to a narrow class of objects....
(iv) It must be psychologically salient for informants. Indices of psychological salience include, among others, (1) a tendency to occur at the beginning of elicited lists of color terms, (2) stability of reference across informants and across occasions of use, and (3) occurrence in the ideolects of all informants ....
(v) The doubtful form should have the same distributional potential as the previously established basic terms. For example, in English, allowing the suffix -ish, for example, reddish, whitish, and greenish are English words, but *aguaish [sic] and *chartreus(e)ish are not.
(vi) Color terms that are also the name of an object characteristically having that color are suspect, for example, gold, silver, and ash. This subsidiary criterion would exclude orange, in English, if it were a doubtful case on the basic criteria (i-iv).
(vii) Recent foeign loan words may be suspect.
(viii) In cases where lexemic status is difficult to assess [see criterion (1)], morphological complexity is given some weight as a secondary criterion. The English term blue-green might be eliminated by this criterion.
**** end of quote ****
Here is a chart of focal colors from twenty languages [ p. 9]. Note that foci for blue and green overlap closely as to hue (the horizontal scale) among languages (so also yellow, orange, brown). The vertical scale is brightness. Saturation is constant in this color set.
Bilingualism may affect color naming (Susan M. Ervin. 1961. "Semantic shift in bilingualism." American Journal of Psychology 74:233-241.)
Loss of color terms "appears rarely, if ever, to happen" . But that does not mean that an acquired term may not be replaced by a synonym or loan word.
The size of the BCT lexicon seems to correlate with cultural complexity and technological development 
See below the mapping of color space for "primitive" languages that have only black, white, & red, or black, white, red, and green:
In the Jalé language, blood is described with their word for "black" (it is a stage 1 language) . Cf. Homer, imitated in Vergil, I think.
Bromley (1967) describes the color develoment of Highland New Guinea languages that use descriptive terms for colors that they don't have basic terms for: "Widely varying descriptive phrases are used for other specific color terms; recurring examples are 'fre leaf' for 'green' and 'cut orchid-fibres' for 'yellow' ...[Bromley 1967: 288, in Bromley M. 1967. "The linguistic relationships of Grand Valley Dani: A lexicostatisticalcClassification." Oceania. 37:286-308.] [B&K 24-25]. Thus we can see that the development of viridis perhaps follows a common pattern among languages.
Many languages lack a word for "brown" .
On Tzeltal green and blue: "Of the forty Tzeltal informants from whom we gathered experimental data, thirty-one indicated that the focal point of yas falls precisely in the area of the spectrum which corresponds to focal English green. In general usage the maximum extension of yas includes greens, blue-greens, blues and some blue-purples. however, when greater specification for yas is requested, many informants restrict the term almost exclusively to greens and some blue-greens. 'blues' and 'purple-blues' are recognized as a distinct area and are designated by a descriptive phrase ... [meaning] 'blackish green' or simply .... 'blackish'. In at least one instance, an informant referred to thie area by the Spanish term azul 'blue'. [32. this reminds me of Servius' description of caeruleus as 'green with black']
!Kung Bushmen of South Africa have a grue word 
Hungarian has a doubling of red terms 
Russian has a doubling of blue terms  but they might involve hyponymy:
"(1) Color terms that 'can be shown on linguistic grounds to be loan words are likely to be more recent additions than native color terms.
(2) Color terms that are analyzable are likely to be more recent additions than unanalyzable terms Analyzability may take five forms:
(i) color terms containing derivational affixes are more recent additions than color terms not containing
(ii) color terms containing more than one stem are more recent addltions than those containing a single stem;
(iii) color terms which contain analyzable stems and/or affixes are more recent addltions than those which
contain unanalyzable stems and/or affixes:
(iv) color terns containing an affix whose gloss is 'color, -colored, color of'. and so on, arc more recent additions than those not containing such an affix:
(v) color terms that are also the names (or contain the names) or objects cbaracteristically having the color in question are more recent additions than color terms which are not (or do not contain) such a name."[37-38]
On these grounds, one can say that luteus, caeruleus, purpureus are later additions because they not only derive from the names of things (2.v) but also show derivational morphology -eus (2.i.)
Western Apache green word describes turquoise 
"The history of many language families, including Indo-European, shows that borrowing a foreign form for a basic color category may serve either to encode a previously uncoded perceptual category or to replace a native form. for example, the French form bleu was probably borrowed from Germanic for a previously uncoded category, while blanc ~ blanche, also of Germanic origin, almost certainly replaced a Romance form." 
The appendix contains a good history of the problem of color terms from the 19th through the 20th century.
There is also a bibliography of color term literature from 1970-1990.