To be honest I never understood the argument that allowing bigger blocks will lead to centralization. First of all, mining node is not even required to keep past blocks - pruned node can mine just fine. Second: the bitcoin blockchain size is almost trivial for even dated PCs (and pruning was implemented ages ago, so it is easy for anyone to run consensus node). Third: while there is a variation depending on transaction sizes, on average doubling size of block only adds one layer to the merkle tree. On top of that the nonce (the thing miners try to find) is added to the header on top of merkle root, the root is computed once per block, so increasing block size does not make mining process slower in any measurable way.
Edit: heh, minutes after sending above I finally found where "the problem" is - the nonce is only 32 bit number, once exhausted the miner has to change the merkle tree somehow, with extra nonce being typically used. Since that extra number(s) is part of coinbase transaction, it changes whole tree, at least on the branch that leads to that transaction. Which means doubling size of block should add on average one extra hash every 4 billion hashes. Doesn't sound that horrible to me, I must be missing something still.