As others have said, what you are trying to do will work, but it will not increase your working memory really.
I would also not really say that this is particularly important. It would be more important if the function and applicability of what you are doing is maximal.
Thinking of working memory in the context of a phonological loop and visuospatial sketchpad is a lot clearer. What you were describing is kind of like using association to buffer a phonological loop (as an example), so that you can recall more with a smaller loop. This is why it would not increase working memory, but effectively would allow you to perform better on a test of memory span.
With regards to spatial-sequence synaesthesia and your approach overall. I know that if I for example perceive a number chunk to be at a certain location it enhances my short-term memory much more than if it is in the same spot. This applies to both the phonological loop and visuospatial sketchpad. The issue I have with this, is in the function. Normally when I stare at a sequence of numbers, I may spot a standard sequence pattern. If I decide to spatially distance these numbers so that I do not process them at once, in this case, while I would perfectly be able to recall each chunk, I may not be able to spot the pattern as readily. That is simply to say that some modifications to working memory operation may cause disadvantages or advantages accordingly. It is really not as simple as improving memory.
With regards to improving working memory, we almost never really exclusively use working memory. For example, all words are long-term memory as are numbers. Therefore, when you use any of these, it is extremely difficult to improve your working memory, because you would simply partly learn a chunk of sorts to reduce the load on your working memory instead. If you think about spaced repetition, it is for example common that some people improve their recall drastically, shortly after having forgotten information, the reason for this is because information is not one unit in the brain, so there would be some partial trace of it which is reactivated. When you attempt to train your working memory, you will constantly be in some sort of chaos against this, because the training intensity changes overtime.
By applicability and function, it is for example sensible that if your approach applies to everything and enhances the functionality of working memory, that it would be beneficial without drawback regardless of the fact that you are not using working memory to as large of an extent. In most cases however such things have more drawbacks than would at first be apparent, but that is really the only issue typically.