If you know the equation of the line that you want to draw, then just
define a variable for your prediction:
predictedResponseTime <- 30 * users ^ 2
If the problem is that you want to fit a line, then you need to call a
loess is a non-parametric model, is isn't appropriate
to use it to make predictions outside of the range of your data.
In this case, a simple (ordinary least squares) linear regression using
lm provides a reasonable fit.
model <- lm(responseTime ~ users)
prediction <- data.frame(users = 1:500)
prediction$responseTime <- predict(model, prediction)
with(prediction, lines(users, responseTime))