installing FME server on a physical or virtual machine should have different performance?

Virtual machines are always tricky because there are so many factors involved, where as on a physical server it's very easy to just look at the Task Manager and you'll get the whole picture.

Have you tried running a generic benchmark tool on both servers to verify that the effective performance is indeed identical?

No, I think the path to your data was different. Either storage or network related.

In our organisation (township Leiden) I have successfully proven that that running fmw in a bubble is way slower.

We were running locally, then the network got virtualised and our performance dropped like a brick.

Now only our profile information is bubbled, fme is installed locally on relevant user machines.

Network issues, security and server performance.

Our Obese Clients are heavy cad machines and much more faster than the servers. (though the servers run on xeon's as do the cad stations)

Even now scheduled processes on server are up to 2-3 slower.

As we can only do this on a couple of server where we have full admin rights.

The log file for each run would be very revealing. Make sure you have log timings turned on in FME Options for maximum information. However... it is still hard to separate out the read-transform-write parts of a translation just from looking at the log. You might consider running each part separately (read only, read+transform) to see where all that time is being spent; or you could add a FeatureHolder transformer after the reader, and after the transformers, and just run the workspace once. Then the single log file will be easier to read.

I do think it's likely to be an issue reading or writing the data. I guess the data was stored on the physical machine? Was it also stored on the virtual machine, or did the virtual machine have to read it from elsewhere? That would make a big difference.

The other thought is that the virtual machine has less temporary file storage. Check FME_TEMP paths and see how much space is available to them. It might be that the virtual machine had to juggle data around on the disk.

When it comes to slow performance, it's often the disk that causes it. Either a lack of memory forces caching to disk (unlikely in this case) or the temporary disk is just not large enough.

I'll add to this that I've seen numerous organisations with a virtualized environment where it was attempted to have everybody run on the same specs. But if you often need to process lots of data with FME your needs are going to be higher than those of your colleagues. One size does not fit all (to be fair, I've also seen this happen with physical hardware)

I've also seen cases (won't name names) where users required a virtual machines with certain specs, along the lines of the FME recommended specs, and ICT gave them the lowest possible specs "until they started to complain"...

Agree, this fits fairly well with my own experiences.