My guess about what caused this is the change in 0.9.1 to match vectors based on pandas index information rather than position, which I think happened in a for-loop over the hues within each category and therefore probably didn't scale well to many categories. I don't know how you would avoid it in 0.11.1 but the good news is that those functions have been rewritten since then and a simple example with the dimensionality that you describe runs quickly (~1 s) now.
Without seeing an example of exactly what you're trying to do it's a little hard to advise, but my guess would be that in most situations with 100 categories you can't really see the jitter/swarming, so you should be able to get away with using scatterplot
...
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…