It's kind of an interesting question and I've heard a lot of folks express different opinions about it with language fro the AIM they claim supports it.
I come down on the side of, if the hold is one that is time based, then the general 1-minute out and then start timing inbound to get as close to 1-minute inbounds as you can is the way to go.
But, if the hold is distance-based (including GPS holds), then you use distance to determine all of the outbounds, including the first. That's answer is not crystal clear from the FAA material but I get there when I think about holds that are =not= 4 nm.
Here's the RNAV 28 into KAPA.
http://aeronav.faa.gov/d-tpp/1101/05715R28.PDF
It has a co-located HILPT and MAPT at DOCKY with 7 nm legs rather than 4. The fact that it's not a standard length tells me that there's gotta be a reason for specifying a leg lenght. My guess is that the reason in this case is traffic separation component, since DOCKY is pretty much right on the approach path to DIA.
That suggests to me that ATC is anticipating that, if you have crossed DOCKY outbound, you're not going to turn inbound for 7 miles. And that you shouldn't (without permission anyway - 7 miles is a long time in a trainer and I generally ask ATC for shorter).