Time spend calling pandoc using bookdown::book_render much greater than using rmarkdownn::render

Hi community,

I've been using bookdown recently. I notice that it takes much more time to render a rmarkdown as a bookdown::gitbook than rendering it as a self-contained html_document using rmarkdown::render. This can be up to three times longer for big projects. This may be expected as it needs to loop over the different sections, but I would like to know if there are bad practices to avoid.

I made a reproductible example with an artificial markdown rendered using two different yaml header, one for bookdown and one for rmarkdown. With this example it takes 35% more time on my machine to render using bookdown.

This consists of the following cell copied/pasted 30 times:

library(dplyr)
numer_of_point <- 2000
num_plots <- 10
data_t <- data.table::data.table(a=seq(1:numer_of_point),b=seq(1:numer_of_point), c=as.character(seq(1:numer_of_point)%%10))
p<- plotly::plot_ly(data=data_t,x=~a,y=~b,color=~c,type = "scatter", mode = "lines")
p
p
p
p
p
p
p
p
p
p
p

I used the following header of the rmarkdown :

---
title: "html_document - test"
author: '`r Sys.getenv("USER")`'
date: '`r Sys.Date()`'
output:
 html_document:
    code_download: true
    code_folding: hide
    toc: true
    number_sections: true
    self_contained: false
    toc_float:
      collapsed: false
      smooth_scroll: false
knitr_params:
 warning: FALSE
 message: FALSE
 out.width: "100%"
 fig.height: 6
 fig.width: 18
---

And the following for the bookdown :

---
title: "bookdown_test"
author: '`r Sys.getenv("USER")`'
date: '`r Sys.Date()`'
site: bookdown::bookdown_site
output:
 bookdown::gitbook:
  code_folding: hide
  split_by: section
  self_contained: false
  config:
   search: false
   info: false
   toc:
    collapse: section
   download: ["html"]
   sharing: false
knitr_params:
 warning: FALSE
 message: FALSE
 out.width: "100%"
 fig.height: 6
 fig.width: 18
---

Any opinion on this topic or idea to reduce the bookdown rendering time would be much appreciated as it is becoming a blocker for big projects (where it takes me up to 40min to render). Thanks!