There appears to several things wrong here, hard to say without looking at the file. Firstly whoever sent you file has sent an incorrect file. As I pointed out CSV is pretty well defined, they haven't followed the rules, and so the file is invalid, and I would return it to them.
Secondly you don't appear to have any documentation, transfer specification or import agreement. Without these your process will be harder to maintain - i.e. data can change next time meaning total re-work - and you have less idea of what the data is. A bad process will always work badly.
Thirdly, this "unbalanced data, which has inconsistent numbers of records in each data update.". It appears, following good Excel usage guides, the data you are using is not even consistent within itself.
Now if I was given this file, and assuming that after many days of complaining, I was told to use this then I would do a pre-scan of the file, i.e. one datastep, read each line as a whole, and have some logic to work out where headers occur. At each of these headers I would write the subsequent lines out to a new file each time. So basically for each of those blocks have a sperate file which does confirm to CSV rules, i.e. one header row, then data separated by commas, Now for each of the those files, write a datastep import, but as they have no standard, and you have no documentation, you would need to look at each file and identify what the data is, how it should be read in etc. Sure you could rely on proc import, and set ranges, but how do you know it has "guessed" the data correctly, how do you know its taken the correct block etc. All in all a lot of problems created by someone not doing things earlier down the line and producing a datafile following standards.
