I prefer the first choice--5 data plots with 3 replications. Because, if you wanna do regression analysis, 5 data plots that is good enough, you can have linear, quadratic relationship etc. Moreover, you wold have opportunity to find out the reliability of your measurement and minimize the error regarding to standard deviation by applying S-hat.

It is better to go for 5 data points with 3 replications, which would reduce the experimental error. if your objective on the central tendency, then both the plans yields almost same answer. However, you may consider the cost and time criteria while taking these samples, whichever is optimum.

It depends on how your variables are (contonious versus categorical) and the statitical approach (e.g. ANOVA vs regression), and most importantly, your questions and the degree of variability of your variables....

It depends on the matter of study too:

In a preliminar laboratory or industrial experiment you may prefer the second option because it lets you to do more combination of factors for looking where it is the optimum, and if you need replications you may add some later without complications.

In an agriculture field experiment if you need replications you have to wait a year before you have a situation comparable to that of the first experiment, Then I would prefer an experiment with replications although it takes more time to find the optimum because it allows to have a check of the lack of fit of the model.

If you want to test the effect of 2+ factors on your response, the second choice is better. If you have only one thing to test, I would do something similar to choice 1. I would do one set of 5 on 3 different days. This will give you an idea about the variability between days.