4. Any other considerations?
Yes. Classic serial ports are ordinarily byte-oriented, ie, one character at a time with quite low latency. USB is inherently packet-oriented, with comparably high-latency. In other words, USB is great for moving a lot of data at once, but it's actually slower at moving tiny amounts of data. If you try to move a lot of data in tiny pieces with USB, you lose - badly.
For just typing on a terminal program talking to a modem or an embedded device, or even uploading files, etc, this is really not a problem, because single doses of the latency are small in human terms, and the USB can easily move blocks of data fast enough keep up with any serial baud rate.
But other protocols which depend on constant back-and-forth of short messages or single characters between a computer and a serial peripheral can slow to a crawl (or even mis-operate) when used with a usb-serial converter. If a computer does something like send a single character, and wait for the peripheral to respond before sending then next, then the character rate is determined by the latency for two entire usb packets - one to send a single character, and one to receive one (it may even be worse - the receive might at the lowest level involve poling). In today's world where USB-serial converters are as or more common than real serial ports, its important to design protocols so that you can send out a lot of data, before needing to wait for a reply. With legacy systems, you may get lucky and find it works fine, it may work but slowly, or it may not work reliably at all.
And even within a given protocol, peripheral, computer, and operating system driver, the design of the client software matters. I once encountered a piece of management software for an embedded device which had been written and tested by a contractor using a real serial port. It turned out to use some variation of .NET serial calls that turned out to be "watch the results crawl in" slow on a laptop with a usb-serial converter. My own previous simple win32 management tool for that device ran without apparently delay on the same laptop with the same converter. Perhaps I wrote better code, but given that the converter was my ordinary test environment, if there had been a similar problem with my code, I would have been lucky enough to find it while there was still time to fix the problem.
Moral of the rant in todays world, anything (hardware or even client software) designed to use a serial port should be tested for usability with usb-serial converters early enough in its development that the design can be changed to fix any issues discovered.