The dead spots in that Poisson distribution are bloody awful but they can always be counted on to appear, as blocks can also always be counted to lump together in streaks. To me the frequency spectrum of that distribution does indeed seem to favour the large pools, because small pools can’t survive the thin parts of the block allocation schedule. It matters little if the sum of all outcomes will still be “fair” if all our delegators have left in the meantime.
Roughly the chances of this happening to you, assuming a geometric average over those 20 epochs of 25% chance of block production, are 0.75^20 = 0.3% … I’ve not had a run literally like this but have been hit with similar anomalies in the block distribution that were equally unlikely.
I suppose the system architects would make the argument that if blocks were given more regularly (i.e. a “lower frequency” probability distribution) it would be more possible to anticipate when individual pools would produce blocks & therefore more possible to disrupt and/or game the system. That’s also why I believe, if you were ever able to attract the attention of one of the IOG scientists about this often-reported issue, that the Ouroboros function would be as unlikely as ever to change. 