Other than speed, what other measures of efficiency might one use in a real-world setting?
Memory
This is probably the most obvious one other than speed. Not only we want to utilize available memory efficiently, we might want to reduce number of memory accesses, avoid leaking memories etc.
One example is learning based approaches often needs to work with very large amount of data. If the dataset is larger than available system RAM, we will have to design our algorithm to be able to run out of core.
Power Consumption
This is particularly relevant when we are developing our algorithm for portable devices like smartphones or smartwatches etc. In such sometimes we want to settle for relatively less power hungry algorithm algorithm that does the job good enough instead of going with more power consuming complex algorithms that produce better results.
One example is anything to do with GPUs. Usually personal computers have GPUs that can consume 75-300 watts, whereas smartphone GPUs have a typical power budget of merely 1 Watt.