I am strongly in favor of increasing block size. Currently there is some momentum for serving scientific data using IPFS. IPFS is actually the perfect use-case for scientific data which is archived and distributed through very clunky repositories, DOIs, etc., all things that are solved very elegantly by IPFS.
One major benefit IPFS has over most repositories is that from these repositories you have to download the data before you can use them (I know of only one that offers S3-compatible object store to access the data) but with IPFS you can actually use the data directly without explicitly downloading them. The issue here is performance and because all of the new cloud ready data formats, such as Zarr, already come chunked, any additional chunking just produces unnecessary overhead.