A study was actually done using most of your assumptions regarding a cash buffer.
http://www.onefpa.org/journal/Pages/Sustainable%20Withdrawal%20Rates%20The%20Historical%20Evidence%20on%20Buffer%20Zone%20Strategies.aspx
TL;DR:
"[T]he use of a buffer zone strategy of any sort—one year, two years, three years, or four years—is more likely than not to leave the investor worse off than if he or she simply set up an investment portfolio with a static asset allocation."
The only advantage of the buffer zone method is psychological:
"the use of a buffer zone may be merited if it will impact one’s investment portfolio choice. To elaborate, it may provide a psychological mechanism to induce clients to accept stock exposure."
The study is looking at the buffer technique, and initially it looked pretty exciting to me - the methodology for withdrawals is "close enough" to what I was proposing.
Unfortunately it looks like the study was set up to cause the buffer method to fail, based on their data set choices. At the very least, the useful data is swamped with crap.
A huge fraction (Okay, 71%) of the data is irrelevant, primarily because they mostly look at withdrawal rates over 4%, which has been pretty much debunked.... everywhere... for any reasonably long withdrawal timeframes. Their range of withdrawal rates used is 3-9% - higher withdrawal rates will pretty much require higher amounts of stock to have even a chance of survival.
Additionally, they are looking at much higher buffer amounts (1-4 years) - only their lowest matches my supposition.
Combine the two (9% and 4 years) and you have a 36% cash position. Of course it fails over 30 years. 36% cash is ridiculous.
...then we have other silly things like having 100% of the "investment" portfolio in bonds.
So, throw out 71% of their data from the overly high to super-high withdrawal rates (5% to 9%)
...we're down to 28% of their data
Throw out 75% of the remaining for high to super-high buffer amounts. (2 years to 4 years)
...we're down to 7% of their data being applicable to the supposition.
Might also throw out the 75% bond and 100% bond scenarios, as I don't know anyone rational advocating this position for 30+ year timeframes.
...we're down to 4% of their data being applicable to the supposition.
They look at both 15 year and 30 year survival rates. 15 year is pretty irrelevant too.
...we're down to 2% of their data being applicable to the supposition.
Yes, they do a fair amount of breakout, but when they (for example) break out 3% SWR, the interesting data is conflated with the other crap data (15 year and 100% bond for example). When they break out 30 years, the interesting data is conflated with the remaining crap data (9% withdrawal rate, 100% bond) et cetera.
Finally, they seriously lowball returns on cash. "Cash" should be earning something close to inflation (longer term CDs, I-bonds) as these instruments can be cashed out with minimal or no penalty - with some restrictions. They instead chose 1-month treasuries, which even under-perform money markets to a serious degree.
5 year CD: 2.25% (BankRate shows 4 banks offering this rate or better)
Money Market: 0.95%
1-month Treasury: 0.02%
Yes, they lowballed returns by 200x
I'll send a nice note to one of the authors and see if they will do a breakout of the more useful data.
That being said, a quick view makes buffers look pretty bad. However, I can't parse the presented data finely enough to actually clearly see what I'm looking for and throw out the ridiculous scenarios.