It occurred to me that Zbigniew Baniewski wrote in gmane.os.netbsd.general:
> 3. I understand, there doesn't exists anything like /dev/ulpa0 - and it
>    cannot (or shouldn't) be created?

No, because there's *ABSOLUTELY* no reason for having it. lpa0 is the 
crippled cousin of lpt0, and only important if you're on a really interrupt
starved machine, that only prints occasionally. That way, the loss of
performance (and extra overhead) caused by the printer driver being
polled might be compensated by the extra interrupt line available for
more important purposes (so PCI cards don't share interrupts anymore)

There are various kinds of printers. Parallel interrupt driven ones
(using the lpt device), parallel polled one (using the lpa device), USB
printers (using the ulpt device), serial printers, and don't forget
the various platform-specific printers (like hppa/ppi). They're all
rather "dumb" devices, all they can do is send data to the printer. What
device you use depends on your hardware (hppa/ppi vs. lp* vs. ulpt),
or specific reasons why you wouldn't want to have an interrupt assigned
to the printer port (lpt vs lpa). 

There is *no* reason to assume that just because lpa and lpt share some
similarities, there should be a -t and -a combo for *every* kind of 
printer port. "ulpa", if such a critter would exist, would be the polling
equivalent of an interrupt-driven "ulpt" - except that "interrupt-driven"
is kind of shady within USB context, as there are no real "interrupts" to
begin with. You couldn't make a difference even if you wanted to. There's
only one way ulpt devices are supposed to work - and that's the way
ulpt uses them.