I think I see what you're saying. I had assumed that once a miner starts solving a block they can't incorporate new tx's that appear after they started. If everyone was solo mining that would make sense. If new transactions are taken into account by pooled mining, then I agree it doesn't help to wait.
If everyone was solo mining it still wouldn't make sense.
There is no such thing as "start solving". Each hash attempt is like a lottery ticket. A winning ticket solves the block, a losing ticket is worthless. Having a million, or billion, or quadrillion losing lottery tickets doesn't mean you are any closer to solving the block then before you started.
Generate block header
So hash 5 trillion tickets
vs
Generate block header
hash 1 trillion tickets
New tx appear on network, create new block header
hash 1 trillion tickets
New tx appear on network, create new block header
hash 1 trillion tickets
New tx appear on network, create new block header
hash 1 trillion tickets
New tx appear on network, create new block header
hash 1 trillion tickets
New tx appear on network, create new block header
in both instances the solo miner has made 5 trillion attempts. The bitcoind is continually updating the working set of the block in background. Anytime the mining software requests work it is on the current working set. If there are lots of paying tx the block reward may be higher, if there are no paying tx then the block reward will just be the subsidy. In all cases it will be at least 25 BTC which is always higher than 0 BTC (not working).