It is true that most CDPs that have 24-bit DACs could not reliably resolve all the gradations of 24-bit. However they almost always do not have to, because the source recording itself is a limitation. The plain oversampling (and upsampling) features just shift the aliasing noise away from the audible band, so that it can be fairly harmlessly filtered off (and at a low cost at that). A lot of the data is just digitally repeated in lower-cost players.
Most that do have 24-bit capable DACs actually still run them at 16-bit anyway, fed at redbook sampling rates (and the DAC internally oversampling 2x, 4x, 8x or more… 88kHz, 176kHz, 352kHz, etc.), instead of the advertised 24 bits and 96kHz or 192kHz sampling (DAC input capability, not actual operation). This is true whether the DAC internally handles the data in native PCM, or uses pulse-proportion modulation (very popular nowadays… mostly on those that claim high multiples of oversampling). It does happen that today’s higher grade DACs are capable of high bit depths and sampling rates, so that can be a justification for the marketing highlight of a “24-bit/192kHz” DAC. Why do DAC makers design their mainstream chips to take 24-bit/192kHz? Because it’s the highest resolution consumer spec (utilized on DVD-A, Bluray and Studio Master Downloads).
There are some that run upsampling, which can do a “non-integer oversample” using a separate chip (and thus changing the DAC's input data feed to 16-bit 96kHz/192kHz). The biggest advantage of using these chips is reclocking, as this tends to reduce both transport and interconnect/interIC jitter (the transport will thus run asynchronously with the DAC). Still fewer use outboard oversampling, which does synchronous integer upsampling before feeding the DAC (because it may reduce native jitter versus having the DAC chip do all the oversampling, and/or it runs the DAC closer to its optimal sampling rate without the losses of upsampling).
Aside from conventional upsampling, some higher end products are now employing stronger processors (Anagram modules with SHARC DSPs are frequently in these devices) that actually convert the data to a very high feed rate with little loss (and bypassing the DAC’s built-in oversampler altogether). These DSPs run at 32-bit or 40-bit internally, and plot the input data points to its internal memory then intelligently outputs a “clean” waveform in the target resolution (upto 24-bit 384kHz and 24-bit 768kHz are fairly common implementations these days). The final resolution is often the maximum capability of the installed DAC chip/s without oversampling. The processing allows for a smoother output (more usable data points derived for the DAC's use), which should theoretically be closer to an analogue waveform. High end outboard DACs often have this functionality as well. When you look at players with this functionality, they tend to be overbuilt and are made with very tight tolerances, so as not to waste the resolution at hand. Sure it may still not be a reliable true 24-bit quality output on the analogue end, but you can tell where a big chunk of your money went.
Not counting the differences on the analogue side (higher end CDPs now often use capacitorless data paths), it is clear that manufacturers are going lengths to extract “better” data from relatively low resolution CDs. It still can’t be as good as a true high resolution format, but as good video upsamplers can enhance video data, so can good audio processing techniques (given that they remain largely non-intrusive).
As for the difference between CDPs and DVDPs, budget CDPs are very much like DVDPs with some different components. A very basic difference is that most CDP transports read at 1x (most DVDPs read CDDA at 1x). DVDPs, often having to run at a higher speed, stop and start reading as their FIFO buffer gets filled and emptied. A problem is that the CD spec (unlike the DVD spec) does not require block-accurate addressing. The address information is frequently pulled out of the subcode channel, so the DVDP may incorrectly identify the exact start of each block (the one just after it stopped reading). This can cause doubled or lost samples. This is a more difficult problem to address than just the plain transport jitter from 1x CDDA-only transports.