I totally agree with you Erik. And that is what spec does'nt say about Ioh max or min, in fact in the old Intel databook you will see details of this characteristic if not in exact values but at least in general trend.
And If I put pullups which will be always sourcing current(> 8 x 0.5mA), instead of an active push pull of the uC, then what I am saving in uC I am blowing it up in the pull up network.
Interestingly the Atmel 89C52 sources approx 240uA into the ULN input and 89C52BP is 20uA. But if I short the pin to ground(that is using the pin as input) then both chips show around 25uA of current. Obviously that extra current in the Atmel chip at output high is the strong pullup at work, missing or weak in case of Philips.
I am sure most of us, if we make an equivalent to an existing product we don't just follow the previous specs, we try to improve, otherwise who needs designers? Cars have got lighter and give more mileage without compromising safety or any other parameter.
That is precisely why in analog design you know whose precision opamp to buy if want the real stuff. But for first time I am experiencing specsmanship in the uC field. A few pieces working may be lucky but a few thousand is not. The transition from 8751H to Atmel 89C52 was smooth to say the least. But now the road looks bumpy ahead I better stay alert.