I think this term was coined in a casual message on GENEALOGY-DNA, but it's never been formally defined.
From context, people seem to mean various things when they use the term.
One is for the persistence of a segment over an "unexpected" number of generations. But this phenomenon is actually to be expected "sometimes". A 10 cM segment will undergo recombination 10% of the time, but the flip side is that it stays intact 90% of the time. If the segment is inherited at all, there's a decent chance that it will remain intact, just like getting heads 3 or 4 or 5 or even more times in a row with a coin toss. It will happen a certain percentage of the time, but people will report it because it seems odd. It is in a sense, but we only see those reports, not the boring ones.
The above would apply to any segment. Another context is where the "stickiness" seems to be a property of the specific segment itself. You might see an unusual number of matches because the segment isn't breaking up as often as expected from the cM value.
The cM unit is empirically determined in a specific sample, so it's an overall average for one group, not a universal measurement. There are some quirky regions of the genome that might not be represented in the sample. For example, there's a somewhat common inversion around the centromere of chromosome 9. If you have that inversion, it's less likely to recombine, and you might match other people with the same inversion even though it would point to a very distant ancestor.
There are also some regions with a few common haplotypes. The HLA region on chromosome 6 is one of those. If there's a common haplotype, you might have inherited it through any of many lines of descent. I discussed some background on the HLA region in my 'Satiable Curiosity column http://www.jogg.info/62/files/SatiableCuriosity.pdf
My general feeling is that the concept of "sticky segment" is overused, but there are some intriguing examples that might repay detailed analysis.