Hmmm ... that's not clear to me.
If you feed 1 byte, 50 bytes, or 5,000 bytes of data to a hash function, does it take more clock cycles to calculate the hash?
The merkel tree is a single hash value of all transactions in a block - Correct?
Does the time to calculate the merkel tree increase with larger argument data (more transactions)?

It might take time to gather more transactions together to feed as arguments into the hash function, but that process has to run regardless, so how much difference can it make to gather 5 transactions versus 20 transaction? I'm not sure but I think we're talking some where between a number of clock cycles or milliseconds where determining the nonce (hence solving the block) is a question of minutes.

If the number of transaction was a significant delay (causing miners to lose the race for the block) I think we would predominately see single transaction blocks. if it's not significant, I would expect to see the maximum that will fit.

Am I missing something here?