> Then there is the mixed IPv6 and IPv4 data that might factor in.
It shouldn't be the problem. The index should separate them on the top level.
> I tried the approach from your benchmark, to try make a more isolated test
> case:
Can you try to isolate it even more by something like this:
select * from routes where route && 'a.b.c.d/e';
It would be easier to debug, if we can reproduce performance
regression like this. It would also be helpful to check where the
time is spent. Maybe "perf" on Linux would help, though I haven't
used it before.