Professor Myhill has conducted a randomised controlled trial into the teaching of grammar involving 744 pupils in 32 classes. Links to her paper and powerpoint presentation are here. This is by far the largest study of its type to have taken place in the UK, and the statistically significant benefits of the grammatical teaching are the most important item of evidence we have of the benefits of teaching grammar.
Grammar was taught in the context of writing narrative fiction, poetry, and constructing an argument. The detail is very clearly set out in Professor Myhill's paper, which should be read by all English teachers, and the results highly impressive, with rouglhly twice the progress made by pupils who were taught grammar in comparison with those who were not.
The limitation is that Professor Myhill found that grammar teaching benefited higher-attaining pupils more than lower-attaining pupils, and that her study did not allow her to investigate this. This is a problem with a randomised trial rather than a focused one. It is highly likely that lower-attaining pupils have problems with different aspects of writing, such as organising their work into sentences, and so may need more focused grammar teaching, with additional practice. Randomised trials make this very difficult, if not impossible, and it is unfortunate that they have become the holy grail of educational research, without any evidence whatever that they are the best way of carrying it out. The very poor study carried out by the originators of this fallacy, discussed in my technical paper, shows that randomisation is only one factor in the design of educational research, and that different types of study provide different types of evidence, all of which may be valuable.