Well, you do have a point. The heat right now is not getting wasted and I can afford the extra cost for electricity, at least for the time being. You are right about the upfront cost, I paid US $50.00 or less for my systems. And I run them with no hard drives, just a live Linux on a USB drive. I also only use one of the two redundant power supplies at a time.
If I could afford more modern and efficient systems, I would upgrade in a heartbeat. I am always on the lookout for anything in the E24xx V2,3 or 4 series, but they have not come into my price range yet. When I see a bargain I will upgrade.
That being said, I find the server architecture to be very dependable as these systems are made to run 24/7 in not the best of conditions. Mine have run with temperatures in the shed from -30F (-34C) to 100F (38C) without a hiccup.
Really wish I could afford one of those dual processor Epyc (Rome) systems though.
Cheers

I have dual boot on my 3900X, so I let it run on windows 10 for a while.

The values are average points/hour/core:

       | Ubuntu | Win10 | %Perf |
| MCM  | 51,75  | 47,47 | 109%  |
| MIP  | 68,87  | 8,27  | 833%  |
| OPN  | 47,03  | 46,16 | 102%  |

So for OPN there is basically zero difference between linux and windows.
For MIP linux seems so much better I actually think there is an issue with the granted BOINC credit in this instance.

These are the results for an older test run on a Xeon 2628lv4:

       | Ubuntu | Win10 | %Perf |
| SCC  | 29,43  | 16,46 | 179%  |
| MCM  | 18,46  | 16,70 | 111%  |
| FAH2 | 16,29  | 18,24 | 89%   |
| ZIKA | 31,16  | 16,79 | 186%  |
| MIP  | 26,26  | 16,98 | 155%  |

I'm running a number of 3770K systems plus one 8700K, all with Win 7 x64, and at times with Debian Linux x64 virtual machines, simultaneously. I always have the total number of threads over the 2 OSes equal to the total for the CPU. The VM infrastructure gives no noticeable speed overhead to the Linux tasks.

The 8700K is at stock turbo 4.3GHz, mem 3000MHzC15, and I picked a 3770K running at 4.34GHz, mem 1600MHzC9, both running OPN on all cores. Yesterday I calculated the mean CPU time for the most recent Valid OPN WUs for each machine under each OS:
. . . . . . . . . . . . . Mean CPU Time (h) per OPN WU
Machine . . . . . . Windows . . . . Linux
8700K . . . . . . . . 2.03 . . . . . . . 2.52 . . . Windows is 23.3% faster than Linux
3770K . . . . . . . . 2.43 . . . . . . . 2.81 . . . Windows is 15.6% faster than Linux
Under Windows, the 8700K was 19.3% faster than the 3770K at comparable clockspeed
Under Linux , . . the 8700K was 11.8% faster than the 3770K at comparable clockspeed

Hope this helps

There has been a frequent recommendation for MIP & ARP to restrict use to less than half the threads available on each machine for different reasons, with other projects taking up the rest. This is not necessary with OPN, SCC or MCM.

Mike