I found your questions are intersting. I think you may:
Question 1: N=2 this is for sure.
question 2: you can't put yt-1 in the estimation of cointegration. What you have found in the other article is definetly wrong.
Question 3: No!
Question 4: Please make sure that y1 x1 x2 are integrated in same order or I(1). You need to check whether there are any possible structural break in the time series. I can say that most of the macro-sereis are bound to struancrual breaks. Pease check with Perron (1997), or Zivot and Andrews (1992) or Perron And V. (1992). If there is structural break, you need to incoparate the sturcturla break into consideration. you may use Gregory and Hansen (1996) test.
question 5: Of course you may any other methods. such as EGtwo steps, Hansan FILM. ARDL etc.
I hope this will assist you in some sense. (P.S. I really want to write in Chinese but my computer have some problem in keyin Chinese charaters. Sorry for that.)