Multiple NVMe devices Interoperation Performance Problems

@All NVMe devices users

I have the opportunity to use different NVMe PCIe x4 v3.0 SSD devices on my ASUS Sabertoothx99 (up to date) machine W10 x64.
They are: Intel i750 400GB, Samsung 950 PRO 512GB, Samsung 960 PRO 1TB.

Doing different configurations I observed severe performance delta depending the way the NVMes devices are sollicited.
I would like to understand why these problems, and if some of yours have also observed similar problems.

The unique Scenario Reference Test is quite simple:
Do a Full System (Source) backup on a local drive (Destination).
The same “System Backup Application” is used for all the tests.
The “System Backup Application” does offer two possible GUI interface: one under W10 x64, the other one under WinRE x64 bootable media (RE means Recovery, and is similar to WinPE).
The Full W10 system backuped Image size is about 38 GB.

All the manufacturer’s NVMe drivers are up to date versions.
The W10 x64 system is the same for all the tests (v1607 Build 14393.576).
The WinRE x64 bootable media image does offer the same W10 x64 NVMe drivers versions (injected via “NTLite” Tool).

Configuration_1:
Source= 950 PRO => Destination= i750
W10 x64 Interface: 800 MB/s max
WinRE x64 Interface: 1180 MB/s max

Configuration_2:
Source= 950 PRO => Destination= 960 PRO
W10 x64 Interface: 350 MB/s max
WinRE x64 Interface: 440 MB/s max

Configuration_3:
Source= i750 => Destination= 960 PRO
W10 x64 interface: 610 MB/s max
WinRE x64 interface: 445 MB/s max

Conclusion:
You can observe significant (and unexpected) delta performances depending of the ‘data flow’.
How to explain the worst performances ?
How to explain the best performances ?