I'm sure this is well-known (think I saw a post about it here), but I can't find the reference. Anyway, this is a clear example of the phenomenon.

The problem is: find 5 hold'em hands of two hole cards each, H1, H2, H3, H4, H5, such that
(a) H1 is a favorite heads-up against each of H2, H3, H4, and H5, but
(b) If all of H1, H2, H3, H4 and H5 go all-in together, then H1 is the least likely to win.

Answer:
Let H1 be K2, H2 be QJ, H3 be 76, H4 be 34, and H5 be 89.

Then if all 5-hands are all-in together, H1 only has 13.5% equity, H2 has 28.2%, H3 has 18.7%, H4 has 17.5% and H5 has 22%. So H1 here is a big underdog.

But headsup, H1 is a 52-48% favorite against H2, 51-49% against H2, 54-46% against H3, and 51-49% against H5.

This still seems paradoxical almost. I wonder if it is related to Simpson's Paradox.

It's only distantly related to Simpson's paradox, it's closer to Condorcet's.

It may seem easier to believe when you consider that 32% of boards have three cards of the same suit. Playing one-on-one, H1 loses to a flush only 7% (against H3 or H4) or 9% (against H2 or H5) of the time. It wins roughly 55% of the time if no one has a flush. That comes out to a slight advantage.

With all four hands, H1 wins 20% of the time if there is no flush. So it's again about even with the other hands. But this is a big disadvantage overall due to the 32% chance of a flush.