Perhaps there is an unwritten rule that this can only be addressed during beta....
Just kidding. Don't take it serious but this is how it can be interpreted as well.

Most important is to get a project into production and fix errors not found during beta.
And my linux servers are assigned to this project despite the fact of higher runtimes.
Nevertheless an improvement in speed is highly appreciated.

I am also seeing this discrepancy. My singe Windows 10 system is seeing impressive run times on FAAH2 units, while the majority of my systems, which all run Linux Mint, are taking substantially more time.

For example, Windows 10 (64 bit) on an i3-4160 cpu is seeing about 11.5 hrs run time. Linux Mint (64 bit) on an i5-2400 cpu, which has a higher performance benchmark, is seeing 15.5 hrs run time.

I'll point a few of my other more powerful Windows systems at FAAH2 and see how they fare as well.

B.t.w. I have observed something similar, but my Windows PC has a faster CPU but indeed it seems quite a lot considering the Linux PC is not a very older generation of Intel CPU.

Indeed I noticed a "Linux vs. Windows" issue during the beta last September.
I've mentioned this issue several times (at least by each bundle of beta WUs).
My summary comment about this issue mentioned that, in my humble opinion, it was not only a question of granted credit but maybe/probably an effective performance problem.
I am operating side by side Phenom II x6, Athlon II x4, and i4770K; currently only the latter host is running Windows 7 Pro SP1 x64. The other hosts are running Ubuntu 14.04 respectively Linux Mint 17.2.
Regarding the CPU performance, the average computation power per core is relatively similar between the system. For this reason, the Linux-Windows issue became rapidly obvious:

• About twice so long run time by Linux hosts: 26-30 hours vs. 11-14 hours for Windows
  
• "Clamped" credit for Linux hosts vs. more "meaningful" credits for the Windows host.

From other projects - e.g. OET1 - we know that Windows-based hosts tend to suffer under a poorly efficient memory management (the memory footprint per WU on Windows systems is much bigger than on Linux systems). It is the reason why Linux hosts usually are able to compute more work per hour than Windows hosts.
Based on all these facts as well as on the member feedbacks during the last 2 days, I think that we are currently experiencing a significant problem with FA@H2.
I am not sure that the trouble is only caused by an inadequate compiler optimization (a such cause would be easy to solve).
The short terms "workaround" would be to avoid FA@H2 for Linux hosts. However it is not a solution for a long terms perspective, since the project needs contribution.
In the past, FA@H was already not terribly well granting comparing to the other projects. The "excuse" was to consider this particular project as an "old" project, maybe suffering under limited optimizations.
Nevertheless, it would be definitively wrong by the launch of a new project - even if it is a phase 2 - to repeat the same failures.
I am still considering that the granted credits do reflect in some way a crunching performance. I correlate the earned points with my electricity bill (5-6 hosts 24/7/365 since several years). I am not really enjoying to waste kWh and money.
Hopefully the problem will be considered seriously and be solved.
Good luck,
Yves

I cannot confirm that, credits are varying under linux as well. My impression is that my Intel CPUs get systematically around 50% more credit per task compared to my AMDs, that is around 580-590 for Intel, and mostly 384.2 on AMD. Maybe there is a bonus system for fast response (like GPUGRID)? On occasion I get ~380 on Intel as well. And then there are these beauties:

FAH2_ avx101121_ 000090_ 0023_ 002_ 1-- 	Valid 	10/2/15 22:36:46 	10/4/15 07:15:35 	24.36 / 25.16 	384.2 / 77.7
FAH2_ avx17558-ls_ 000031_ 0010_ 002_ 1-- 	Valid 	10/3/15 15:04:15 	10/4/15 21:16:54 	20.96 / 21.19 	570.3 / 188.8

No idea what happened here. I think they are worth a closer look. But I think that somebody already stated during beta that the credit system for FAH2 is completely broken. Getting less than 78 credits for more than 24 hours of computing on an Intel Core i7-3930K is not very realistic...

But for me, the runtimes are far more of an issue. I hate to think that I am wasting CPU resources due to poorly optimized binaries. That doesn't serve anybody.