Hmmm. Characterising low-light performance is actually much harder than it might appear.
The DxO rankings are quite good given the conditions under which they test. IIRC, the sports score is just the maximum (interpolated) ISO at which a sensor produces an image at a given Signal-to-Noise Ratio (SNR). Keep in mind that DxO scores are (still?) comparing images downsampled to 8 MP, which they use as their representative "print size". This is a much better idea than it sounds at first, because it allows you to compare full-frame and APS-C sensors in a meaningful way.
Here are my tips:
1. Anything can look good at web-sized images, even ISO 256000
2. Do not pixel-peep (at full sensor resolution) when doing sensor noise comparisons
3. Resize all images to the same resolution before comparing them (this is the "print size" normalization that DxO does). This could mean scaling all images down to "print size", or scaling all images down to the size of lowest-resolution sensor in your comparison.
If you follow those tips, then a full-frame sensor will give you roughly 1 stop better performance (i.e., apparent noise in print size image at ISO 6400 on FF will look the same as ISO 3200 on APS-C). This is because of the larger sensor in the FF camera capturing more photons in the same amount of time (which reduces the "photon shot noise" --- a physical phenomenon that cannot be circumvented). The sensor surface area ratio (FF area / APS-C area) about 2.2, which is neat because that is roughly equal to the 1 stop improvement we see.
With a modern Sony/Nikon sensor the actual pixel count does not harm noise performance, i.e., a 16 MP sensor and a 24 MP sensor will produce comparable noise at print size. This is not necessarily true for many other sensors, so check.
There are other subtle issues too. For example, an FF sensor at 24 MP (D610) will still have larger photosites (pixels) than a 24 MP APS-C sensor (D7100). This impacts the number of photoelectrons that can be stored per pixel, i.e., the larger pixel will be less prone to saturation (overexposure). In the field, this means the FF sensor is better at handling highlights without clipping.
If you want to form your own (real-world) opinion, download the test shots on dpreview (they usually have the raw files available for download). Develop them with NR as you normally would, but be sure to compare them at your desired print size equivalent (which might be roughly A4 size, for example, meaning 8 MP). If you use the dpreview tool to compare studio shots, keep in mind that you are pixel peeping, i.e., a lower resolution sensor will inherently appear to have less noise, but it will also have less detail --- this makes it hard to compare low-light performance objectively because you are conflating two different aspects that react differently to NR algorithms.
How about the D610? It's going to offer several stops better high ISO performance than what you have, as well as double the resolution, much better dynamic range, better image quality... And it can be yours for around R25k. I may have missed it, but do you need a lens as well? If so, the D7100 with 18-140 lens can be had for under R20k and will fit about halfway between what you have and the D610.
