Cannibalized intro point circuits are now 4 hops, instead of 3 (HS-side)
The fix of legacy/trac#16260 (moved) makes hidden services establish a 4-hop introduction circuit, if they cannibalized during introduction circuit construction.
This was done mainly to prevent circuit fingerprinting according to the latest USENIX paper, but it's unclear whether making only this change helps with anything.
Hence, the performance penalty of 4-hops might be more important here.
Another question, is how often do introduction circuits get cannibalized? It might be that it often happens, since this step is done in the very beginning when there are a few internal circuits lying around. But I might be very wrong.