The host is Windows 7 x64
32 bit version of Win2003 guest hangs during boot if more than 4GB of RAM selected (even if PAE is off), kernel debug through serial port shows it break in ntkrpamp.exe (ie NT kernel).
32 bit version of Win2003 R2 guest boots but is very unstable (corrupt files and hangs spontaneously)
32 bit version of Win2008 guest works fine with either PAE on or off.
Tested on VMware Player version 12.1 and 7.1.3
But even if you select less than 4GB of RAM, the setup of latest VMware Tools 10.0.5 crash Win2003 during the HGFS driver install phase. It looks like you do not even test 2003 guests.
It was solved by changing mem.hotadd option to FALSE in vmx file. Why on earth do you patch kernel on the fly to implement some questionable feature and not even test it?
Which edition of WIndows Server 2003 (including service pack and release) you're using ? Take in mind that if using Standard edition, since you're using a 32 bits version, there is a limitation of 4GB, see: Memory Limits for Windows and Windows Server Releases (Windows)
I tested these guests:
Windows 2003 Enterprise Edition 5.2 build 3790 Service Pack 2 installed.
Windows 2003 R2 Enterprise Edition (can't see build number now) also with Service Pack 2 installed.
Enterprise edition or standard edition doesn't mean much because I tried to run it with /nopae switch and even /maxmem=2048 switch. It just crashes if a VM is set to more than 4GB even if kernel does not map all the ram. It worth noting that safe mode works just fine.
VMware Tools setup crash on Win 2003 but installs fine on R2 version.
It was solved by changing mem.hotadd option to FALSE in vmx file. Why on earth do you patch kernel on the fly to implement some questionable feature and not even test it?