If you want a supercalculator that can quickly do the operations on huge matrices and arrays, sure, use the 50G. I seldom need that though. In fact, most of what I do on a calculator doesn't particularly need a lot of speed. I have a 71B as well as the 41cx, and although the 71 is far faster and more powerful, I seldom use it. The 41 is more practical as a calculator.
Although changes in my work mean I don't do this nearly as much as I used to, I also want the machine to be able to interface to other equipment-- and I don't mean just a printer or mass-storage device, or even a PC. I want to connect it to programmable signal generators, signal-routing boxes, power supplies, power meters, multimeters, analyzers, etc.; and the 41, using the HPIL and Extended I/O modules could easily do that. Using the HP82169A HPIL-to-HPIB interface converter, there have been literally thousands of models of lab instruments that can be interfaced to an HP-41, 15 at a time. A man I work with is working on making his 50G able to do that, apparently by building up something he saw in a publication to turn the 50G's serial port into real RS-232 so that he could connect an RS232-to-HPIB (IEEE-488) interface, then write the software to control it, starting at the low-level commands. He has worked on it off and on for many months. What a mess, to do something the 41 did with ease! In my early years with the 41, the small company I was working for started making an audio product that was going to require automated testing. I set up some instruments and used the 41 as the controller, which was much easier to take home at night than a laptop. If I took a commuter train home, it would have been easy to work on things on the train. I didn't realize how quickly this was going to spiral out of control, and it soon got to where test operators were using my 41 most of the day to test product. I got a bigger HP controller with a 68000 processor and coverted things over, but the speed only increased about 70%, because both computers spent a lot of time waiting for filters to settle and readings to come back from the equipment. IOW, the 41 was not slowing things down much. The first two million dollars of that product however were tested with my 41 and a program that was about a dozen pages.
I'm not fond of menus. I do want a well-written paper manual like the 41cx's two volume one, and like the 71B's Owner's Manual and Reference Manual. Someone gave me a 50G and I've gone through the User's Manual, and I have to say the manual is absolutely terrible (to put it mildly!) compared to my 41's and 71's manuals. I have not started on the online manual yet.
I don't need a bigger display. The biggest projects I've done with the 41 required minimal human I/O. I have the 24-line, 80-column HPIL video interface, and I haven't used it in years. What I have been doing in recent years when I write a program is to do it on the PC so I can use the hi-res monitor, have lots of comments, and put more than one instruction on a line, then when I'm happy with it, key it into the 41. Most of my 41 programs these days are only a couple hundred bytes, so it doesn't take long to key them in.
I also don't need graphics, although it would be nice if the 41 had a dot-matrix display so it could do all the lower-case and special characters better.
Are there improvements I would like in the 41? Sure. But the exposure I've had to the 50G so far makes it much harder for me to cozy up to.