Would you elaborate please what you mean by 'it doesn't scale'?
The time taken to process a block doesn't increase linearly with the size, it's quadratic.
If you double the blocksize, you multiply the time taken to process it by 4.
It gets very slow very quickly.
Only for those transactions with have multiple inputs or outputs. So in the real world the validation time for a doubling in blocksize is going to be somewhere between 2 - 4 times longer. The more fragmented the UTXO set, the more it will tend towards the upper bound. And since even using big UTXO's to spend non UTXO amounts requires 1 input and 2 outputs, it is likely that it will be far higher than the lower bound.
The question is how much modern CPU's can cope with it, versus the miners risking of getting a block orphaned due to the validation time delay in releasing a block. So far miners have gone from 0.25MB, to 0.5MB, to 0.75MB to 1MB and coped. Now they can't go to 1.25MB because of an original anti-spam limit intention.
After reading above comment, this is about CPU validation time of transactions that are included in a block, and has nothing to do with hashing power.
Thank you, I can more-or-less follow that and it makes sense to me. I'm hoping, though I can't entirely trust myself on this, that that is more because I follow the reasoning in your logic than that I'm agreeing because we're in agreement of the harmfulness of the current block-size limit!