The answer is, it depends Seriously what would be slow, delivery of data to clients, bittorrent shows distributed p2p is not slow.
Tying together requests with who, why etc. would be slower (and very difficult) but do we even want that?
The bottom line is that a local sql server will be faster at SQL and that is OK, but it wont be scale-able. Facebook / Amazon etc found that out, so scaled to distributed locally protected networks. So distributed and partially decentralised networks are what they use. They though will interconnect these with high speed bandwidth channels so it will be fast.
Then you consider they ignore security to a great extent, but surround their network with firewalls etc. and lots of administrators. So this ignore security also speeds things up. In addition they own the hardware, so spec it appropriately and also manage it (network management agents per machine reporting health etc.). This allows them to ignore failover protections or self healing properties, again speeds it up.
However to do this, they need something we don’t think any company or person should have and that is control over access to information. In fact usually with T&Cs they own the information you give them.
With all those caveats they can be faster, but so can a car if you remove the brakes and seat-belts