Hi, Bennet. I'm not 100% sure I'm answering the right question, but I'll give it a go. My first assumption is you're not referring to the colors you can set for "Tag Group Management" under your profile. I find that feature really handy for flagging triangulation groups once I've identified ones I'm highly confident in. But in that case, you're setting the color and then adding kits to the groups manually. So we're not talking about that.
If it's the Tier 1 "Matching Segment Search" (actually titled "GEDmatch DNA Segment Search" once you get to it), I think that seemingly cryptic answer you found was correct...if not stated very well: "The change in color indicates a change in matching. The colors themselves don't mean anything."
If you choose to include graphics in the display, you get little colored bands positioned approximately where the specific segment would be relative to the beginning and end of the chromosome. Here's a link to one of Kitty Cooper's blog posts written shortly after the feature was introduced. Scroll down to the second screen capture. Below it, Kitty notes: "The colors in the graphic section just indicate where the logical breaks are in the overlaps, they are not otherwise significant." I believe the key is "logical breaks."
How exactly GEDmatch determines that, I don't know. My suspicion is that it simply has to do with sequential proximity of the SNPs tested. Since all our current autosomal DNA tests make assumptions about a segment's start and end points, and its unbroken continuity, based solely on the SNPs sampled (on a given segment you might have 700 SNPs that match contiguously but they're spread out over 1 million base pairs) some math modeling goes into determining the base pair numbers of the start and end points. In most cases, the length in actual base pairs and the evaluation in centiMorgans is never completely accurate. The reason is that only those 700 SNPs were identified; each SNP aligns with a specific base pair, or reference cluster; but matching base pairs may well continue fore and aft...or maybe even break in the middle somewhere. We're assuming a sequence of 700 SNPs that match identically mean that all 1 million base pairs the SNPs encompass also match identically. Until full-genome testing becomes affordable, we'll never really know for certain.
And the density of the SNPs tested vary all over the place, not just from one chromosome to another but with significant differences along the same chromosome. On some little stretches of a chromosome, the SNPs will be packed in there like sardines, while other stretches are virtual SNP deserts where the nearest neighbor SNP is a long way up the road.
All that to say the seeming overlaps in the coloring of the segments may have more to do with the clustering of identical SNPs than the actual overlap of the base pairs. For example, a segment might begin in a SNP dense area, then trail off into a SNP desert. A second segment might begin near the end of that desert, but continue through another SNP-dense area. In terms of the matching SNPs, the conclusion might be that there is a "logical break" between the two segments in that SNP desert; that the two segments likely belong in two different groups because they don't share the same nearby areas of high SNP density.
What I can say in support of Kitty's statement that "they are not otherwise significant" is that the color scheme is a spectrum based on the segment's location along the chromosome, from start of the chromosome to the end. The color groupings aren't otherwise tied to specific test kits in any way. They have (unfortunately) nothing to do with matrilineal/patrilineal lines of ancestry because none of the genealogy testing companies can supply the raw data in a way that distinguishes mother from father. Dang it.
Last, the "Matching Segment Search" can't be set below a threshold of 5cM and 500 SNPs. Also kinda unfortunate, but you could imagine what would happen to their servers if people started asking for 200 SNPs and 2cM thresholds and to show all chromosomes. But 5cM is still works.
I'd do it one chromosome at a time, but it's still a rather useful way of seeing if there are localized pile-up regions that might not be phylogenetic (shared by very large segments of the population), but are specific to your own haplotype. It isn't that these are necessarily false positives, but if there are a whole boatload of matches sharing the same smallish segment, there's a reasonable possibility that the segment is deeply ancestral...meaning that it's stuck around over many generations. I'd then consider those segments to be dubious in genealogical matching. With time and effort in building triangulation groups with 10 or 20 people in them, a widely-shared segment can become meaningful. But without that substantiation, it's possible that segment can't be matched to a MRCA because the couple is simply too many generations back. These small segments widely shared among matches are perhaps the greatest pitfall of triangulation attempts; they're why simply finding two other people who share a small segment with you will never an accurate triangulation make.