Regarding the top hat radials, they're more for aesthetics than functional, but the original 1st generation, 23ch P500s had the top hat radials bent down at a 45° angle, whereas the later & last Hy-gain production run of the 40ch P500s had them sticking straight out at a true horizontal,
...but I have a copy of the assembly paperwork for the 23ch, circa 1975 and the drawings picture BOTH - LoL.
The following is to show what my Eznec modeling would predict regarding the issues you noted for the SP 500.
I have several old manuals for the SP500/525 made by HyGain, Telex, including the New HyGain version. They show the radiator varying in length with the older manuals showing a longer radiator at 273" inches, and over the years the radiator got shorter and shorter to today's model at 264" inches...with other lengths at different points in time.
These models also show the Top Hat radials slanted down in some versions and being flat in others. A couple of manuals show the antenna image with a slanted Top Hat, while in the same manual showing the assembly diagram indicating the Top Hat radials are flat. Apparently the Top Hat idea was not consistently applied over time. I also agree this difference likely makes little difference.
Below I made two Eznec models, one with a slanted Top Hat vs. another model with a Horizontal Top Hat. NB, if we compare these it shows us very little difference in match and the gain, but like you suggest the difference would not be detectable just using your radio.
I was surprised to see the difference in the maximum RF angles between these two models however. The slanted radial show us a drop in the maximum gain angle of 2* degrees and a small reduction in maximum gain value indicated. Here we also see a small change in match. So NB, there is a difference.
I also think this top hat design may serve more than window dressing too. I can't prove it, but I have heard that making the highest point of a vertical antenna tip larger in size...helps drain static and possibly helps minimize lightning strikes as well.
I did find out that if I removed the Top Hat and adjust the radiator length to show the same reactance and adjust the segment count so the AGT is good...then the model does show a bit less gain. I can't explain this for sure, but IMO the model has improved a bit as a result of removing the TH and thus the gain due to error has been removed...thus the gain shows a little reduction. Maybe I'll post this model later if anybody is interested.
Note: Check the images in the 2nd group of PDF files below. The first two images of the antenna dimensions were sent me years ago. They were marked as 23 channel and 40 channel, but IMO he got them backwards, because the 40 channel exhibit shows the radiator longer than the 23 channel.