How to load and then patch in 16 bit environments?

How does it "not work" ?

From the code you posted above, I would most likely guess that your "loader" has not resized his own memory and will be using all memory up to 640 KB which means there is simply no room for the other program to be loaded.

Even if you fix this, you don't set up DS and ES before jumping to CS:IP.

And how do you expect to get back to your "mov ah,4ch / int 21h" code after the jump to the other program? It's not like you would be calling something which would return with a "retf".

And how would you like to "SCASB" yourself to the location you want to patch? If you go for INT 21/AX=4B01, you will have the location you want to patch relative to CS:IP of the loaded program. If you're going with INT21/AX=4B00, DOS will load, execute and unload the program without giving you even a chance of patching something.

You have to understand that DOS had no constant memory management, no support for multi tasking and no support for IPC (expect the 4F0 area). Have you ever seen any DOS memory patchers? They all hooked some interrupt vectors and watched the call address to match some specific values.