The attention tax is a good instrument for a book with one primary investment and a poor one for a book with several. Applied to Middlemarch it produces confident nonsense: that Eliot should have cut Lydgate, that Tolstoy should have chosen one family, that Dickens should have decided between Chancery and Chesney Wold. Those books are not tolerated exceptions. Their threads are not deductions from a main plot, and Empson’s argument says why: the relation between plots produces a meaning that neither carries, and a pruned novel cannot have it. The present preference for a single tightly held line is largely a fact about the length of books and the market that sets it, passed on as craft advice.
The tax is still worth applying, because it is a stress test rather than a description. A thread that survives it has a function its writer can name, and a thread that fails may still belong to a multi-plot novel, provided the writer says so and accepts the duties that come with the form: declaring it early, letting the relations carry the argument, converging or rhyming at the end. In this book’s judgement the commonest structural fault in an ambitious manuscript is not a second plot but a book that has half-declared itself a multi-plot novel while keeping half a hierarchy, so that the reader cannot tell whether a thread is a co-plot to be weighed or a subplot to be tolerated. The tests cannot answer that question for the writer. What they can do is make its answer unavoidable, and the answer is a decision about what kind of book this is, which no count of pages will make.
See also: the Reference coda, on the digressions of Les Misérables and Moby-Dick and what a pruned novel loses.