A big reason is that the two ways are almost the same, so it makes sense to choose one instead of supporting both.

The HTTP approach has two advantages over the DNS approach:
1. it's easier to add other metadata to the pubkeys, like we do today in some cases;
2. it's 10000x easier to support dynamic name resolution, for providers that have multiple users, each with its own name and pubkey.

The first NIP-05 draft used DNS records, but then I realized that it was inferior.