Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
The other posts have well covered the details, but I just want to emphasis the key to understanding Portfast is it's a STP port that skips initial listening and learning while blocking the port (as described by Nandan).
Documentation about using this setting for hosts is because this is what Portfast was provided for, to allows hosts to join the network without waiting for STP to unblock the port. Just a single edge host isn't going to create a loop and earlier hosts used to have difficulity with the delay between getting link-up and being able to actual "talk" to anything (as original Etherent was shared. I.e. there was no port delay with 10Base2 or 10Base5 or 10BaseT shared hubs. (I.e. Link-up, host is hot on the network.)
So, Portfast allows a host to come on the network a bit quicker because you're telling the port it shouldn't need to worry about loops.
Interestingly, though, Portfast doesn't disable STP, again it just skips initial link-up listening and learning while blocking. As noted in the other posts, if Portfast detects a STP loop it will still block the port. The problem, though, is your infrastructure get take quite a hit until (and if) a looped network is blocked by STP. Without Portfast, a loop isn't allowed (all this assumes we have BPDUs, etc.) because we wait to unblock a port when we know it is safe to do so. With Portfast the port is unblocked and we might have to deal with a loop flood and then try to block the port.
With rapid-STP, Portfast also doesn't send TCNs with link up/down status change while port is still in Portfast (edge) mode.
If this makes sense, then you can see why you could connect multiple downstream hosts via a hub or "dumb" switch (or switch with STP disenabled or filtered) and a Portfast doesn't care. I.e. it really doesn't care about hosts.
On your question about conserving bandwidth by filtering out BPDUs, on typical LAN circuits, and switches, the "insurance" of having STP active is well worth the slight performance hit to block an inadvertent loop. In theory if your topology is loop free (i.e. no redundancy) you don't need STP, but all it takes is one "oops" and your whole network can crash and burn. Features like STP, even in non-redundant L2 topologies, is an important safeguard and one reason to use "smart" or manageable switches that support it rather than $99 48 port non-managed switches.
