your characterizing Windows as "retarded" for not distinguishing between 750 char/s and the much faster network, was illogical.
There are two parts to that.
The first part is that the network log-in source can be grouped as an infinite number of terminals--lots of connections--so a per-connection rate limit is useless; thus all network service log-in (caveat: Active Directory handles console log-ins... over network) must be grouped as one thing to be effective. Console log-ins are separate so that a network attack can't function as a DOS; as well, the risk is mitigated because you can't enter passwords fast enough for any use.
The second part is that a console brute-force is slow. Your concern about what amounts to typing really, really fast (i.e. programmed HID plugged into USB) isn't a real concern because of password complexity. It's not that passwords are necessarily that complex; it's that a password which isn't complex enough can be readily brute forced under strong password policies like "3 passwords per minute", it just takes a week or two.
You dismiss the possibility that weak passwords are used, so that hardware password attacks are dismissable, but at the same time address the problem that these same non-weak passwords aren't strong enough to withstand network password attacks without lock-outs?
No, I dismiss the possibility that short lock-out intervals help with weak passwords.
You can attack 129,600 passwords per 30 days if you have a 3 failure per minute policy. Basic English 1250 extends out to about 5000 words with conjugations and domain language (medical, legal, whatever) for most people. Weak passwords in the traditional complexity scheme are like "rainman" becomes "Rainman1", so 100,000 attempts has a fair chance of getting it eventually. That's within the realm of a hardware keyboard typist. Common policy is 60 or 90 day retention, which increases the risk into strong viability; while public kiosks are too visible for a multi-hour console log-in attack, which makes these attacks less viable even at high rates.
Complex passwords reach 10^14 theoretically, and four-word passwords reach 10^16. Reasonable rate limits of 20 attempts per minute carry this out to hundreds of thousands of years. A human can type barely that fast. Remember the original argument:
Windows does stupid shit like lock the local console if you set up rate-limit log-in...when logging in through the Microsoft log-in manager.
If the attacker tries to log in over RDP or telnet or such, and locks the account, the actual console log-in box no longer works. That's dumb, because no attacker can possibly brute force the password through that, unless the password is laughably weak--in which case, as stated above, the rate limiting doesn't actually help.
tl;dr: I can lock you out of your server by constantly trying to log into your server, so you can't apply patches anymore. Then I hack it on Tuesday.