This is indeed an odd inconsistency that should lead one to question how reliable is firecalc (or other tools). Anomalies like this happen all the time in modeling and is a sign that you have too little data and things can't be as precise as they appear.
Well, the inconsistency with Firecalc is pretty easy to explain. Let's say you put in a 50 year timespan. Then, it's only going to look at 50 year periods of 1964-2014 (or does it have to wait until 2014 is complete?), 1963-2013, 1962-2012, and so on. It simply can't model 1965-2015, 1966-2016, and so on, because those timeframes aren't complete yet.
However, if you put in a 40 year timeframe, then it will show 1974-2014, 1973-2013, etc. Well, the later half of the 60's were not good years, as the economy pretty much peaked in 1965 and stocks went stagnant, and inflation started to tick up. And the 70's were downright tumultuous.
One way to sort of predict a longer retirement though, would be to plug in, say, your 30 or 40 year timeframe, but then look at where most of the plots end. If you have a 100% chance of success after, say, 30 years, but 5 of those plots end close to the zero mark, then there's a good chance that those might dip below zero after a few more years. Well, there are 114 30 year cycles in the Firecalc program. So, discount those 5 as failures, just to be safe, and your success rate is really 95.6%.
Just for kicks, I re-ran my own numbers, for what I consider my most likely scenario, which is retiring in 2020 at the age of 50. Here's the variables. Currently 44, planning it out to age 100, or the year 2070. I know that's probably overkill, but you never know. I have one Grandfather who will be 100 if he makes it to October 24 of this year.
Anyway, I have a 96.6% chance of success, for this 56 year cycle, ending in 2070. But, here's what happens when I shorten it...
96.8%: end at 2065 (51 years)
93.9%: end at 2060 (46 years)
94.2%: end at 2055 (41 years)
96.3%: end at 2050 (36 years)
98.2%: end at 2045 (31 years)
Now, for that final, 31 year scenario, only 2 out of 113 cycles failed, so it seems good at first. However, looking at the plots, it looks like 3 or 4 are getting pretty close to zero at the end of 31 years, so there's a good chance they'd fail not long after. And overall, it looks like maybe 10-15 have my portfolio at less than the starting balance, so it's very possible they could keep trending downward.