Tom, you are supposed to say bad things about FileMaker on the FileMaker forum, not good things! You are so out of fashion.
I suspect you got cracking VM from your cloud provider - check the specs. Who did you go with?
Also, are the layouts in question optimised for WebDirect?
The way WD works (in part) is to render the entire layout and then minor changes (say entry into a field) are ONLY updated in that location/position on the page. In other words, the entire screen is NOT refreshed. Look at AJAX examples/tutorials to see what I mean. A conditional (or cascading) value list set would be the same. ONLY the lists change, not the entire page.
OK, this is not the whole story, but I hope it helps you understand. In many ways the same thing happens in FMP & FMGo, but perhaps there are enough of the elements in WD that make it appear to be screen-refreshed faster?
And yes, the browser can cache content and if a website is well designed, it can be significantly faster. So heed nicolai's question. OPTIMIZED fmpro layout can also be faster (in Web Direct, FMGo or even server/client)
I used the sample file so I am sure the layouts are optimized for WebD. I will poke at it some more tomorrow.
It is on a server with GoDaddy. The entire server is 4x 6core E5-2630L v2 2.4. The cpu power is shared. I usually get 4 cores of solid performance with burst of 6 cores when I need it on my similar Linux servers. I wonder why anyone would get a dedicated server from them when the VPS seems to have more cpu resources available. I read a few people say FMS won't run with GoDaddy, but that may only be true for the cheap shared web servers. Supposedly mirrored across a few data centers, but I'm not for certain.
I think we have touched on this before, but WD still receives an entirely undeserved bad press on performance.
When I gave users the option of using a deskspaceApp based system on FMGo or WD last summer they opted for the WD option because they found it significantly faster than the FMGo client. It is also generally faster in my testing than the FMP client.
In your Server Load testing with dsBenchmark - you may have noticed this difference when using WD?
As Beverly has observed this is because WD is a virtual client session running on server, hence the amount of network activity is much less i.e. just the user input going up and the screen coming back. A normal client session involves FMP or FMGo controlling what Server does, the client session running on the device, interacting with server down whatever network you are using, hence it is very much slower.
By having the local client session offload specific tasks to Server using PSOS you do get better performance as you will have seen, much better in complex cases like creating or deleting a record, but provided you have 32mb of spare RAM on Server used neither by the OS nor within the FMS cache, then WD will always be significantly faster _provided_ you have written simple efficient code suitable for that use and have ensured that no local styles have crept into your theme.
Best regards, Nick